Non-uniform scale pointer meter reading identification method

Abstract

The invention provides a non-uniform scale pointer meter reading identification method, and relates to the technical field of pointer meter reading, and the method comprises the steps: S1, obtaining a to-be-read instrument panel image based on a visual module, and carrying out the preprocessing of the to-be-read instrument panel image; s2, performing rectangular expansion processing on the to-be-read instrument image to obtain a rectangular expansion graph; and S3, fusing the auxiliary pictures with the standard scales to form a reading diagram. According to the method, the decimal part X in the reading graph is estimated based on binary separation, the image is separated, then the area where the pointer is located is recognized, the process is very simple, the used recognition models are consistent for any pointer meter reading with non-uniform scales, and it is guaranteed that model training has enough annotation data; most importantly, the training data label of the model is only used for judging whether the pointer exists in the image area or not, manual labeling is very convenient, the operation mechanism of the model is simplified, and the pointer meter reading with non-uniform scales can be accurately read.

Description

Technical Field

[0001] The present invention relates to the technical field of pointer meter reading, and specifically provides a method for identifying the reading of a pointer meter with uneven scales. Background Art

[0002] As an important device for industrial field monitoring, pointer-type instruments are widely used in fields such as power systems, chemical production, and mechanical control. Traditional pointer meter reading identification mainly relies on manual visual judgment, which has problems such as low efficiency, poor real-time performance, and being easily affected by subjective factors. Especially in complex industrial environments, there are potential safety hazards, and it is impossible for humans to reach the site where the instrument is set to observe, etc.

[0003] Since the reading of a pointer meter with uneven scales cannot obtain the final data result by simply reading the length data and then performing calculations, the reading of a pointer meter with uneven scales overly relies on model training, requires a large amount of labeled data for model training, and has problems such as high computational resource consumption and insufficient generalization ability. Therefore, the present invention provides a method for identifying the reading of a pointer meter with uneven scales. Summary of the Invention

[0004] Aiming at the deficiencies of the prior art, the present invention provides a method for identifying the reading of a pointer meter with uneven scales, which solves the problems that the reading of a pointer meter with uneven scales overly relies on model training, requires a large amount of labeled data for model training, and has problems such as high computational resource consumption and insufficient generalization ability.

[0005] To achieve the above objectives, the present invention is realized through the following technical solutions:

[0006] A method for identifying the reading of a pointer meter with uneven scales includes the following steps:

[0007] S1. Obtain an image of the meter to be read based on a vision module, and preprocess the image of the meter dial to be read;

[0008] S2. Perform rectangular expansion processing on the image of the meter to be read to obtain a histogram expansion diagram;

[0009] S3. Integrate an auxiliary picture with standard scales to form a reading picture;

[0010] S4. Read the integer part T in the reading picture;

[0011] S5. Estimate the decimal part X in the reading picture based on binary segmentation;

[0012] S6. Add the integer part T and the decimal part X, and output the result as the final data.

[0013] Preferably, the image of the instrument to be read is an image obtained with the lens perpendicular to the dial plane, uniform illumination, and a resolution of not less than 1280×720.

[0014] Preferably, the preprocessing of the image of the instrument to be read includes: color space conversion to reduce computational complexity, highlighting shape features, Gaussian blur and median filtering to eliminate noise, and histogram equalization to enhance global contrast.

[0015] Preferably, in step S5, the specific method for estimating the fractional part in the reading map based on binary partitioning includes:

[0016] S501: Perform central partitioning within the unit area where the pointer is located. Record "0" on the left side, and the data length is represented by a; record "1" on the right side, and the data length is represented by b.

[0017] S502: Select the area where the pointer is located in the two areas formed after the central partitioning in step S501 as the second unit area, and repeat step S501 within the second unit area.

[0018] S503: Similarly, execute step S502 N - 1 times to obtain the path binary sequence (S1, S2,..., S N ), S i ∈{0, 1};

[0019] S504: Calculate the fractional part X;

[0020] ①;

[0021] In formula ①, if the N th digit in the path binary sequence (S1, S2,..., S ) of the area where the pointer is located is 0, then ; if the N th digit in the path binary sequence (S1, S2,..., S ) of the area where the pointer is located is 1, then ;

[0022] In formula ①, N represents the total number of the path binary sequence (S1, S2,..., S N ), represents the specific value of the N th serial number in the path binary sequence (S1, S2,..., S ) of the area where the pointer is located, a represents the data length when recording "0" on the left side, b represents the data length when recording "1" on the left side, and a + b = 1.

[0023] Preferably, the number of layers for binary separation estimation is 6 - 10 layers. The higher the number of layers for binary separation estimation, the higher the accuracy of the fractional part X.

[0024] Preferably, in step S501, central separation is performed within the unit area where the pointer is located. The left side is recorded as "0" with the data length represented by a; the right side is recorded as "1" with the data length represented by b, which specifically includes:

[0025] a. Identify the unit area where the pointer is located, then convert the image into a single-channel grayscale image, and symmetrically crop the center of the unit area where the pointer is located into two first unit areas;

[0026] b. Based on the YOLOv5 model, identify the first unit area with a pointer within the two first unit areas.

[0027] Preferably, in step S2, the image of the meter to be read is processed by rectangular expansion to obtain a histogram expansion diagram, which specifically includes: using contour analysis to confirm the center of the image of the meter to be read, setting the radius of the image of the meter to be read, and based on the center and radius of the image of the meter to be read, expanding the image of the meter to be read into a rectangle to obtain a histogram expansion diagram.

[0028] Preferably, in step S3, the auxiliary picture with standard scales is fused to form a reading picture, which specifically includes:

[0029] Identify the scale zero point of the histogram expansion diagram, then align the scale zero point of the histogram expansion diagram with the scale zero point of the auxiliary picture with standard scales, and scale the histogram expansion diagram / the auxiliary picture with standard scales in the length direction so that the histogram expansion diagram and the auxiliary picture with standard scales are fused to form a reading picture.

[0030] Preferably, it further includes: error correction for the final data result output:

[0031] Process multiple images of the meter to be read obtained continuously, and exclude those with obvious jumps in the final data result;

[0032] Then calculate the average value of multiple final data results that meet the requirements.

[0033] The present invention provides a method for identifying the reading of a pointer meter with uneven scales. It has the following beneficial effects:

[0034] 1. In the present invention, by processing the image of the meter to be read through rectangular expansion to obtain a histogram expansion diagram, and fusing the histogram expansion diagram with the auxiliary picture with standard scales to form a reading picture, most of the influencing factors in the image of the meter to be read are isolated in this process, reducing the interference with the subsequent process of reading the integer part T in the reading picture and estimating the fractional part X in the reading picture based on binary separation.

[0035] 2. In the present invention, the fractional part X in the estimated reading graph is separated based on binary, and the image is separated, and then the area where the pointer is located is identified. This process is a very simple process, and for the pointer table readings with any uneven scales, the recognition model used is the same, ensuring that there is sufficient labeled data for model training. Most importantly, the training data label of this model is only to judge whether there is a pointer in the image area, and it is very convenient to perform manual annotation (assist model annotation), which simplifies the operation mechanism of the model and realizes accurate reading of the pointer table readings with uneven scales. BRIEF DESCRIPTION OF THE DRAWINGS

[0036] Figure 1 is a flowchart of a method for identifying the reading of a pointer table with uneven scales proposed by the present invention;

[0037] Figure 2 is a schematic diagram of binary separation of a method for identifying the reading of a pointer table with uneven scales proposed by the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0038] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present invention.

[0039] Embodiment 1:

[0040] As Figure 1 shown, the embodiment of the present invention provides a method for identifying the reading of a pointer table with uneven scales, including the following steps:

[0041] S1. Obtain the image of the instrument to be read based on the vision module. It is necessary to ensure that the image of the instrument to be read is obtained with the lens perpendicular to the dial plane, uniform illumination, and a resolution not lower than 1280×720, so as to avoid affecting the subsequent reading due to image quality problems such as visual misalignment and pixel blurring in the obtained image of the instrument to be read. And preprocess the image of the instrument panel to be read, use color space conversion to reduce the calculation complexity and highlight the shape features; use Gaussian blur and median filtering to eliminate noise, and use histogram equalization to enhance the global contrast to grayscale the image of the instrument to be read.

[0042] For example, use a high-speed camera with more than 2 million pixels and automatic fill light or a FLIR BFS-PGE-50S5C model camera with automatic exposure (AE) + region of interest exposure (ROI), etc.

[0043] S2. Perform a rectangular expansion process on the image of the meter to be read to obtain a histogram expansion diagram. Comparing with reading the value of a circular meter, it is simpler to read the value on the histogram expansion diagram. Use contour analysis to confirm the center of the circle of the image of the meter to be read, and set the radius of the image of the meter to be read (when reading data for a certain known pointer-type meter, the radius of the image of the meter to be read can be set; of course, the radius of the image of the meter to be read can also be determined based on contour analysis). Based on the center and radius of the image of the meter to be read, expand the image of the meter to be read into a rectangle to obtain a histogram expansion diagram.

[0044] S3. Integrate the auxiliary picture with standard scales to form a reading diagram. Identify the zero point of the scale of the histogram expansion diagram, and then align the zero point of the scale of the histogram expansion diagram with the zero point of the scale of the auxiliary picture with standard scales. Scale the histogram expansion diagram / the auxiliary picture with standard scales in the length direction so that the histogram expansion diagram and the auxiliary picture with standard scales are integrated to form a reading diagram. In the integrated reading diagram, the scale is clear, and there are clear numerical values of the integer part T on the scale. Integrate the pointer position in the histogram expansion diagram onto the auxiliary picture with standard scales. Based on the text reading, the integer part T in the reading diagram can be quickly read, and it provides a clearer picture for subsequent estimating the decimal part X in the reading diagram based on binary segmentation, ensuring that it can be binary-segmented multiple times.

[0045] In the auxiliary picture with standard scales, there is a central dividing line in each unit area, and the data lengths on both sides of the central dividing line are marked, which is prepared for subsequent performing the estimation of the decimal part X in the reading diagram based on binary segmentation; another way is: store the data length representations a and b on both sides of the central division corresponding to the next unit area under the integer part T in the model database. After the integer part T is read, obtain the data length representations a and b on both sides of the central division corresponding to the unit area where the pointer is located according to the integer part T.

[0046] S4. Read the integer part T in the reading diagram, which includes two processes. One is to confirm the unit area where the pointer is located, and according to the unit area where the pointer is located, read the value marked near the left end of the unit area where the pointer is located.

[0047] S5. Estimate the decimal part X in the reading diagram based on binary segmentation. During the binary segmentation estimation process, the model's recognition of the image only involves multiple judgments of the area where the pointer is located, and its judgment method has unity and standardization. After the relevant parameters of the trained model are adjusted, it can read the readings of other pointer meters with any uneven scales, and the data labels used for training this model are also very easy to label manually. Compared with the traditional method of using the readings of pointer meters with uneven scales read manually as labels, this application only needs to make a selection and judgment, so that more training data can be obtained.

[0048] S6. Add the integer part T and the decimal part X and output the result as the final data result.

[0049] S7. Error correction for the output of the final data result:

[0050] Process multiple images of the instrument to be read obtained continuously, exclude the final data results with obvious jumps, and then calculate the average value of multiple final data results that meet the requirements.

[0051] In the present invention, by performing rectangular expansion processing on the image of the instrument to be read, a histogram expansion diagram is obtained, and the histogram expansion diagram is fused with an auxiliary picture with standard scales to form a reading diagram. This process isolates most of the influencing factors (such as shadows, scale defects, misalignments, etc.) in the image of the instrument to be read, reducing interference with the subsequent process of reading the integer part T in the reading diagram and estimating the decimal part X in the reading diagram based on binary separation. Moreover, the present invention creatively proposes to estimate the decimal part X in the reading diagram based on binary separation, separate the image, and then identify the area where the pointer is located. This process is a very simple process, and for any pointer table reading with uneven scales, the recognition model used is the same, ensuring that there is sufficient labeled data for model training; most importantly, the training data label of this model is only to judge whether there is a pointer in the image area, and it is very convenient for manual annotation (assistant model annotation), simplifying the operation mechanism of the model and achieving accurate reading of the pointer table reading with uneven scales.

[0052] When the data length a on the left side and the data length on the right side after central separation are both 0.5, it represents a pointer table with uniform scales; that is to say, applying the technical solution disclosed in this application to the reading process of a pointer table with uniform scales should also belong to the conventional application of this technology.

[0053] In one embodiment, the specific steps of estimating the decimal part in the reading diagram based on binary separation in step S5 include:

[0054] S501. Perform central separation within the unit area where the pointer is located, record "0" on the left side, and the data length is represented as a; record "1" on the right side, and the data length is represented as b.

[0055] The sources of the left data length a and the right data length b can be as follows: In the auxiliary picture with standard scale, there is a central dividing line in each unit area, and the data lengths on both sides of the central dividing line are marked, and the data read and labeled by the text reading model; Another way is: Store the data lengths represented by a and the data lengths represented by b on both sides of the central division corresponding to the next unit area under the integer part T in the model database. After the integer part T is read, obtain the data lengths represented by a and the data lengths represented by b on both sides of the central division corresponding to the unit area where the pointer is located according to the integer part T.

[0056] S502. Select the area where the pointer is located in the two areas formed after the central division in step S501 and define it as the second unit area, and repeat step S501 in the second unit area;

[0057] S503. Similarly, execute step S502 N - 1 times to obtain the path binary sequence (S1, S2,..., S N ) of the area where the pointer is located, S i ∈{0, 1};

[0058] S504. Calculate the fractional part X;

[0059] ①;

[0060] In formula ①, if the N th digit in the path binary sequence (S1, S2,..., S ) of the area where the pointer is located is 0, then ; if the N th digit in the path binary sequence (S1, S2,..., S ) of the area where the pointer is located is 1, then ;

[0061] In formula ①, N represents the total number of the path binary sequence (S1, S2,..., S N ) of the area where the pointer is located, represents the specific value of the N th serial number in the path binary sequence (S1, S2,..., S ) of the area where the pointer is located, a represents the data length when the left side is recorded as "0", b represents the data length when the left side is recorded as "1", and a + b = 1.

[0062] For example, take a = 1 / 3, b = 2 / 3, N = 2;

[0063] The path binary sequence of the area where the pointer is located is (1, 1), X = 1 / 3 + 1 / 3 * 2 / 3 = 0.5556;

[0064] The path binary sequence (1, 0) of the area where the pointer is located, X = 0.3333.

[0065] The number of layers of the general binary separation estimation is 6 - 10 layers (the total number N of the path binary sequences (S1, S2,..., S N ) of the area where the pointer is located is 6 - 10). The higher the number of layers of the binary separation estimation, the higher the accuracy of the decimal part X.

[0066] For example, when the number of layers of the binary separation estimation is 2 layers, its last time is equivalent to dividing into 4 areas (2 2 ) When the number of layers of the binary separation estimation is 10 layers, its last time is equivalent to dividing into 1024 areas (2 10 ) Since this method is an estimation, it doesn't make much sense if the number of layers of the binary separation estimation exceeds 10 layers.

[0067] In one embodiment, step S501 performs central separation within the unit area where the pointer is located. Record "0" on the left side, and the data length is represented by a; record "1" on the right side, and the data length is represented by b. Specifically, it includes:

[0068] a. Identify the unit area where the pointer is located, then convert the image into a single-channel grayscale image, and symmetrically crop the center of the unit area where the pointer is located into two first unit areas;

[0069] b. Based on the YOLOv5 model, identify the second unit areas where the pointer exists within the two first unit areas.

[0070] Although the embodiments of the present invention have been shown and described, for those of ordinary skill in the art, it can be understood that various changes, modifications, substitutions, and variations can be made to these embodiments without departing from the principles and spirit of the present invention. The scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. A method for identifying the reading of an analog pointer watch with uneven graduations, characterized in that, It includes the following steps: S1. Obtain the image of the meter to be read based on the vision module, and preprocess the image of the meter to be read; S2. Perform rectangular expansion processing on the image of the meter to be read to obtain a histogram expansion diagram; S3. Integrate the auxiliary picture with standard scales to form a reading diagram; S4. Read the integer part T in the reading diagram; S5. Estimate the decimal part X in the reading diagram based on binary segmentation; S6. Add the integer part T and the decimal part X, and output the result as the final data.

2. The method for identifying the reading of a pointer watch with uneven graduations according to claim 1, characterized in that: The image of the meter to be read is an image obtained when the lens is perpendicular to the dial plane, the illumination is uniform, and the resolution is not less than 1280×720.

3. The method for identifying the reading of a pointer meter with uneven graduations according to claim 2, characterized in that, Preprocessing the image of the meter to be read includes: converting the color space to reduce the calculation complexity, highlighting the shape features, performing Gaussian blur and median filtering to eliminate noise, and performing histogram equalization to enhance the global contrast.

4. A method for identifying the reading of a pointer watch with uneven graduations according to claim 1, characterized in that, In step S5, estimating the decimal part X in the reading diagram based on binary segmentation specifically includes: S501. Perform central segmentation within the unit area where the pointer is located. Record "0" on the left side, and the data length is represented by a; record "1" on the right side, and the data length is represented by b; S502. Select the area where the pointer is located in the two areas formed after the central segmentation in step S501 as the second unit area, and repeat step S501 within the second unit area; S503. Similarly, perform step S502 N - 1 times to obtain the path binary sequence (S1, S2,..., S N ) in the area where the pointer is located, and S i ∈{0, 1}; S504. Calculate the decimal part X; ①； In formula ①, if in the path binary sequence (S1, S2,..., S N ) of the area where the pointer is located, the -th digit is 0, then ; if in the path binary sequence (S1, S2,..., S N ) of the area where the pointer is located, the -th digit is 1, then ; In formula ①, N represents the total number of path binary sequences (S1, S2,..., S N ) in the area where the pointer is located, represents the specific value with the serial number N in the path binary sequence (S1, S2,..., S ) in the area where the pointer is located. a represents the data length when recorded as "0" on the left side, b represents the data length when recorded as "1" on the left side, and a + b = 1.

5. A method for identifying the reading of a pointer watch with uneven graduations according to claim 4, characterized in that: The number of layers of binary segmentation estimation is 6 - 10 layers. The higher the number of layers of binary segmentation estimation, the higher the accuracy of the decimal part X.

6. A method for identifying the reading of a pointer table with uneven graduations according to claim 4, characterized in that, In step S501, perform central segmentation within the unit area where the pointer is located. Record "0" on the left side, and the data length is represented by a; Recording "1" on the right side and representing the data length by b specifically includes: a. Identify the unit area where the pointer is located, then convert the image into a single-channel grayscale image, and symmetrically crop the center of the unit area where the pointer is located into two first unit areas; b. Based on the YOLOv5 model, identify the first unit area with a pointer in the two first unit areas.

7. A method for identifying the reading of a pointer meter with uneven graduations according to claim 1, characterized in that In step S2, performing rectangular expansion processing on the image of the meter to be read to obtain a histogram expansion diagram specifically includes: using contour analysis to confirm the center of the image of the meter to be read, setting the radius of the image of the meter to be read, and based on the center and radius of the image of the meter to be read, expand the image of the meter to be read into a rectangle to obtain a histogram expansion diagram.

8. A method for identifying the reading of a pointer watch with uneven graduations according to claim 1, characterized in that, In step S3, integrating the auxiliary picture with standard scales to form a reading diagram specifically includes: Identify the scale zero point of the histogram expansion diagram, then align the scale zero point of the histogram expansion diagram with the scale zero point of the auxiliary picture with standard scales, and scale the histogram expansion diagram / auxiliary picture with standard scales in the length direction so that the histogram expansion diagram and the auxiliary picture with standard scales are integrated to form a reading diagram.

9. A method for identifying the reading of a pointer watch with uneven graduations according to claim 1, characterized in that, It also includes: Error correction for the output of the final data result: Process multiple images of the meter to be read obtained continuously, and exclude those with obvious jumps in the final data result; Then calculate the average value of multiple final data results that meet the requirements.

Images